---
license: apache-2.0
base_model: distilbert-base-uncased
tags:
- text-classification
- email-classification
- distilbert
- pytorch
- transformers
language:
- en
datasets:
- custom
widget:
- text: "🎉 FLASH SALE! 50% OFF everything! Limited time offer - Don't miss out!"
  example_title: "Promotions Email"
- text: "Your verification code is: 123456. This code will expire in 10 minutes."
  example_title: "Verification Code"
- text: "Welcome to our technical discussion forum! Join developers worldwide."
  example_title: "Forum Discussion"
- text: "CONGRATULATIONS! You've won $1,000,000! Click here to claim your prize!"
  example_title: "Spam Email"
metrics:
- accuracy
model-index:
- name: email-classifier-distilbert-base-uncased
  results:
  - task:
      type: text-classification
      name: Email Classification
    metrics:
    - type: accuracy
      value: 1.0
      name: Accuracy
---

# Email Classifier - DistilBERT Base Uncased

## Model Description

This model is a fine-tuned version of `distilbert-base-uncased` for email classification tasks. It can classify emails into 6 different categories:

- **forum**: Forum discussions and technical conversations
- **promotions**: Marketing emails and promotional content
- **social_media**: Social media notifications and updates
- **spam**: Spam and unwanted emails
- **updates**: System updates and subscription renewals
- **verify_code**: Verification codes and authentication emails

## Model Performance

The model achieves **100% accuracy** on the test dataset with perfect precision, recall, and F1-scores across all categories.

### Test Results

| Category     | Precision | Recall | F1-Score | Support |
|--------------|-----------|--------|----------|---------|
| forum        | 1.000     | 1.000  | 1.000    | 3       |
| promotions   | 1.000     | 1.000  | 1.000    | 3       |
| social_media | 1.000     | 1.000  | 1.000    | 3       |
| spam         | 1.000     | 1.000  | 1.000    | 3       |
| updates      | 1.000     | 1.000  | 1.000    | 3       |
| verify_code  | 1.000     | 1.000  | 1.000    | 3       |
| **Overall**  | **1.000** | **1.000** | **1.000** | **18** |

## Usage

```python
from transformers import DistilBertTokenizer, DistilBertForSequenceClassification
import torch

# Load model and tokenizer
model_name = "your-username/email-classifier-distilbert-base-uncased"
tokenizer = DistilBertTokenizer.from_pretrained(model_name)
model = DistilBertForSequenceClassification.from_pretrained(model_name)

# Example usage
def classify_email(text):
    inputs = tokenizer(text, return_tensors="pt", max_length=384, padding=True, truncation=True)
    
    with torch.no_grad():
        outputs = model(**inputs)
        predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
        predicted_class_id = predictions.argmax().item()
        confidence = predictions[0][predicted_class_id].item()
    
    labels = ["forum", "promotions", "social_media", "spam", "updates", "verify_code"]
    return labels[predicted_class_id], confidence

# Test the model
email_text = "Your verification code is 567890"
category, confidence = classify_email(email_text)
print(f"Category: {category}, Confidence: {confidence:.3f}")
```

## Training Details

### Model Architecture
- **Base Model**: distilbert-base-uncased
- **Model Type**: DistilBERT for Sequence Classification
- **Parameters**: ~66M parameters
- **Max Sequence Length**: 384 tokens
- **Number of Classes**: 6

### Training Configuration
- **Vocabulary Size**: 30,522
- **Hidden Size**: 768
- **Number of Attention Heads**: 12
- **Number of Hidden Layers**: 6
- **Intermediate Size**: 3,072
- **Activation Function**: GELU
- **Dropout**: 0.1
- **Attention Dropout**: 0.1

## Intended Use

This model is designed for email classification tasks and can be used to:
- Automatically sort incoming emails
- Filter spam and promotional content
- Organize email communications
- Prioritize important messages like verification codes

## Limitations

- The model is trained on English emails only
- Performance may vary on emails with different formatting or domains
- The test dataset is relatively small (18 samples)
- May need retraining for domain-specific email patterns

## Ethical Considerations

- This model should not be used to violate privacy or read personal emails without consent
- Consider data protection regulations when implementing in production
- Be aware of potential biases in email classification

## Citation

If you use this model, please consider citing:

```bibtex
@misc{email-classifier-distilbert,
  title={Email Classifier DistilBERT Base Uncased},
  author={Your Name},
  year={2024},
  howpublished={\url{https://huggingface.co/your-username/email-classifier-distilbert-base-uncased}}
}
```

## Contact

For questions or issues, please contact [your-email@example.com]